Book Review
The Numerati
Reviewed by Jeffrey Shallit
The Numerati
Stephen Baker
Houghton Mifflin Co., 2008
US$26.00, 256 pages
ISBN-13: 978-0618784608
Is it possible for a nonmathematician to write
both accurately and entertainingly about a math-
ematical topic while still conveying something
nontrivial about the mathematics? The answer is
yes, but good examples are rare. Constance Reid
did the trick with her book about Hilbert, and, to
a lesser extent, with her book about Courant, but
she had the advantage of having Julia Robinson
for a sister. And of course Martin Gardner, who
had little formal mathematical training, wrote the
“Mathematical Games” column of Scientific Amer-
ican for many years, and introduced the beauty
of mathematics to many young readers, including
this reviewer.
Stephen Baker, the author of The Numerati, is,
unfortunately, no Martin Gardner. By numerati (the
word apparently first appeared in a 1990 review of
a British art exhibit, written by Doron Swade) Baker
means the kind of people who, were they working
in the financial industry, would be called “quants”:
people with very strong mathematical and com-
puter skills who can analyze real-world problems.
While “quants” study financial markets and build
mathematical models, Baker’s numerati analyze
large volumes of data collected electronically, in
order to make predictions about human behavior
in a variety of spheres: voting, employment, con-
sumption, crime, illness, blogging, and marriage.
Each of these activities gets a chapter devoted to
it, in which Baker interviews several people who
Jeffrey Shallit is professor of computer science at the
University of Waterloo, Ontario, Canada. His email ad-
dress is shallit@cs.uwaterloo.ca.
October 2009 Notices of the AMS
analyze the relevant data and try to come up with
marketable conclusions. “[T]hese mathematicians
and computer scientists,” Baker intones sternly,
“are in a position to rule the information of our
lives.”
In a book whose subject is data, equations,
and mathematical models, Baker is surprisingly
shy about presenting any actual mathematics. Or
perhaps it is not so surprising. Steven Hawking
once wrote, “Someone told me that each equation
I included in the book would halve the sales. I
therefore resolved not to have any equations at
all. In the end, however, I did put in one equa-
tion, Einstein’s famous equation E = mc 2 . I hope
that this will not scare off half of my potential
readers.” [2] Baker has taken Hawking one equation
further.
Baker’s approach is almost entirely anecdotal.
All told, he interviews about two dozen of the
numerati, ranging from IBM’s Samer Takriti to Ya-
hoo’s head of research, Prabhakar Raghavan, and
asks them some not-very-revealing questions. We
learn very little about their personalities and even
less about what it is they do on a day-to-day basis.
Some of the anecdotes are, admittedly, inter-
esting. I particularly enjoyed the plan to “put
a wireless computer on half a million cows in
Kansas”; with the data collected, researchers hope
to determine what behavior patterns of cows are
correlated with higher-quality meat. But some are
not so interesting. Baker opens with a puzzle: why
do people who rent romantic movies online also
tend to click on an ad for rental cars, much more
than the average user? The answer, when it comes,
is not that surprising: lovers of romantic movies
were attracted by the ads that promoted weekend
“escapes”.
There is very little in The Numerati to interest
the professional or amateur mathematician; this
is the kind of book that a business executive
1109
 might buy in an airport bookstore, hoping to learn
something about mathematical modeling and the
Internet—but I imagine even the business execu-
tive will find insufficient novelty in Baker’s modest
survey. There’s just not enough detail provided to
tell the reader very much about the main subject:
the models and algorithms that extract meaning
from large volumes of data.
As an example, consider this passage: “If one
of Raghavan’s scientists gives an imprecise com-
puter command while trawling through Yahoo’s
data, he can send the company’s servers whirring
madly through the noise for days on end. But a
timely tweak in these instructions can speed up
the hunt by a factor of 30,000. That reduces a
24-hour process to about three seconds. His point
is that people with the right smarts can summon
meaning from the nearly bottomless sea of data.
It’s not easy, but they can find us there.”
Reading this, I can only wonder, what is an “im-
precise computer command”? Does the passage
concern a new breakthrough at Yahoo in search
optimization, or something obvious that every
undergraduate computer science student learns,
such as binary search? Baker just doesn’t give
enough detail to decide.
Baker emphasizes that the volume of data col-
lected by the numerati requires new techniques,
but he doesn’t really explain why. It would have
been nice to read something along these lines: if
we are working with small data sets, with hun-
dreds or thousands of items, we can afford to
use algorithms that run in linear, O(n log n), or
even quadratic time. But, as my colleague Alex
López-Ortiz has noted [3], when you are dealing
with 230 or even 240 data points, the log factor is
the difference between a query that completes in
a second and one that completes in half a minute.
Too often Baker relies on clichés. Over and
over, we are told that the goal of the numerati is
to “turn us into dizzying combinations of num-
bers” (p. 13), to “turn IBM’s workers into numbers”
(p. 20), and that they will view people as “boiled
down to numbers” (p. 23) or “represented as a
series of numbers” (p. 35). Of these, only the last
is accurate. Sometimes, though, Baker says we are
actually equations: “each of us [is] represented by
scores of equations” (p. 42); “I had ... no clue as
to what kind of equation I would become” (p. 99).
This, even metaphorically, seems incorrect. Peo-
ple might be represented by numbers, and their
relationships might be governed by equations, but
it makes little sense to claim that an individual’s
attributes are represented by an equation.
Although most of his account is accurate—as
far as it goes—Baker does get some of the histo-
ry wrong. He claims, for example, that “Google’s
breakthrough, which transformed a simple search
engine into a media giant, was the discovery that
our queries—the words we type when we hunt for
1110 Notices of the AMS
Web pages—are of immense value to advertisers”.
This is incorrect. The site Goto.com allowed adver-
tisers to bid on search results as early as February
1998, two years before Google did so. Google’s
original noteworthy accomplishment—and the one
that made it the search engine of choice—was its
new algorithm, called PageRank, for deciding what
Web pages provide good matches for a query.
PageRank represented the Web as a directed
graph. Nodes are pages, and there’s a directed
edge from page A to page B if A links to B. In its
simplest form, PageRank assigned a weight W to
the edge (A, B) with
number of links from B to A
W = .
total number of pages that B links to
The resulting square matrix, called the “link ma-
trix”, is column stochastic and has an eigenvalue
of 1. The associated eigenvector, if it is unique
and suitably normalized, gives the “rank” or im-
portance of each page. (There is now more actual
mathematics in this review than in all 244 pages
of Baker’s book.) To make this idea work well in
practice, we need uniqueness of the eigenvector
and a fast way to calculate it, so the mathematical
story doesn’t end here. But even in its infancy,
PageRank helped Google give much better results
than other search engines—so good that Google’s
home page cockily offers an option labeled “I’m
Feeling Lucky”, where only a single search result,
the top one, is revealed—that it quickly became
the search engine of choice. Although Google’s
search engine has since moved far past PageRank,
a mathematically savvy writer could have easily
summarized these elementary ideas, or at least
referred to the paper of Bryan and Leise [1].
Even when a simple geometric diagram would
have enlightened the reader, Baker refuses to pro-
vide it. In talking with Mark Steitz, a Democratic
consultant, he describes a “simplex triangle” that
represents voters in an election. Each voter is
represented by a point with two coordinates that
represent (a) the likelihood of favoring one party
over another and (b) the likelihood of actually
going to the polls in any election. “Steitz draws a
vertical line up the triangle, a so-called isoquant.
Each voter along this line is of equal value, he
says.” Although I imagine every reader of this
review could produce the diagram Steitz has in
mind, one picture here would be worth more than
a hundred words.
In the chapter on politics, Baker discusses the
difficulty of obtaining good data on who people
are likely to vote for. Because of this, “proxies”
are used; if you bought a Volvo and shop at Trad-
er Joe’s, you might be more likely to vote for a
Democrat than someone who’s an NRA member
and drives a pickup truck. Geographical proxies
can be good predictors, too, but Baker’s account
Volume 56, Number 9
is superficial compared to others, such as Michael Ultimately, I did not find The Numerati a very
Weiss’s The Clustering of America [5]. satisfying account of its subject. I wanted more
The contrast between this book and some re- insight—something that Baker, with his nonmath-
lated ones published recently is startling. For ematical background, could not provide. Perhaps
example, Emanuel Derman’s My Life as a Quant I am unfair in criticizing Stephen Baker for not
[4] is a memoir of the author’s career as a physi- writing the book I would have wanted to read. The
cist, computer programmer, and financial wizard. problem is, I don’t think he wrote the book that
Along the way, Derman provides portraits of most people would have wanted to read.
Tsung-Dao Lee, the physicist who co-discovered
the asymmetry of the weak interaction with C. N. References
Yang, and Fischer Black, co-creator of the Black- [1] Kurt Bryan and Tanya Leise, The
Scholes equation for the value of an option. Here $25,000,000,000 eigenvector: The linear alge-
bra behind Google, SIAM Review 48 (2006),
is Derman on T. D. Lee:
569–581.
... every speaker felt compelled to [2] Steven Hawking, A Brief History of Time, Bantam
focus on him; as they spoke, their Books, 1998.
eyes fixated only on him, and he [3] Alejandro López-Ortiz, Algorithmic foundations
let no statement he did not fully of the Internet, Combinatorial and Algorithmic
Aspects of Networking, Lecture Notes in Comput-
agree with pass him by. No mat-
er Science, Vol. 3405, Springer, Berlin, 2005, pp.
ter who lectured at the seminar,
155–158.
T. D. concentrated intensely on [4] Emanuel Derman, My Life as a Quant: Reflections
their argument, and interrupted at on Physics and Finance, Wiley, 2004.
the first instant something was not [5] Michael J. Weiss, The Clustering of America, Tilden
satisfactory. At times he broke in Press, 1988.
on the initial sentence of the talk,
refusing to let a speaker proceed
until the point was clarified. Some-
times clarification never came; I
once witnessed the humiliation of
a visting postdoc who was forced
to defend the first sentence he ut-
tered for the entire hour and a half
allowed for his seminar.
Derman’s writing is witty, insightful, and mov-
ing; his prose is eloquent, and accurately captures
the joys and sorrows of doing research. Derman’s
book is not filled with equations, either, but he
uses diagrams effectively to make his points, and
describes, in a clear if nontechnical way, some
of the ideas that excited him in physics and fi-
nance. As someone who has actually worked in
mathematics, physics, and finance, Derman writes
with an authority and insight that Baker cannot
approach.
Very little of The Numerati is devoted to an
analysis of the ethical and privacy concerns that
data collection raises. Although Baker briefly dis-
cusses one way of hiding from the numerati—an
initiative called Attention Trust—he says almost
nothing about technologies for cryptography and
anonymity. Modern cryptography, which is strong-
ly mathematically based, offers us the hope that
many of our transactions can take place veiled
from the prying eyes of the numerati. And anony-
mous Web-surfing, based (for example) on tech-
nology from anonymizer.com or the Tor project,
can prevent data collectors from linking online
behavior with the specific person who is doing the
surfing.

October 2009 Notices of the AMS 1111